Non-adherence and loss to follow-up in a clinical trial threatens the validity of conclusions about the intervention. The Botswana Isoniazid Preventive Therapy (IPT) clinical trial was conducted between November 2004 and July 2009. It was a double-blinded, randomized, placebo-controlled clinical trial to determine whether isoniazid taken daily for 36 months was more effective in protecting against tuberculosis (TB) in HIV-infected adults compared to the standard-of-care in which isoniazid was taken daily for six months. The researchers reported that participants receiving 36 months of IPT had half the risk of TB compared to participants receiving the 6-month regimen [1]. During the conduct of the trial, a sub-study was conducted to assess the causes of non-adherence.

While IPT reduces the incidence of TB disease among HIV and TB co-infected individuals [2], adherence may be difficult because of the long duration of prophylaxis. Typically, patients with chronic conditions have poorer medication adherence than those with acute conditions [3]. It is important to understand factors associated with non-adherence so that high adherence can be maintained or low adherence improved [3]–[6] since adherence to effective treatment improves health outcomes [7], [8].

Non-adherence to medication regimens is observed in clinical trials and is common under routine program conditions. Participants may wish to stop taking the medication or cease returning for clinic appointments. In order to better understand what factors affected non-adherence in our study, we conducted a sub-study in which we identified two distinct groups of non-adherent participants:participants who refused to take the study medication but agreed to return for study visits and participants who were lost to follow-up. Demographic characteristics and other risk factors of these two groups were compared against those who remained adherent in the Botswana IPT clinical trial. As an earlier analysis of trial data found that being on antiretroviral therapy (ART) was associated with better IPT adherence [9], we included ART use as a risk factor.

Between November 2004 and July 2006, 1995 participants were enrolled in the clinical trial. Overall, the clinical trial had a very good adherence rate during the 3 years of follow-up with 78% of participants adherent to at least 80% of their scheduled visits, more than three-quarters of randomly selected participants having detectable isoniazid metabolites in their urine (Table 1) and 91% having reached a known study endpoint by 36 months of follow-up.

Adherence and retention in clinical trials are important issues because participant non-adherence and loss to follow-up can compromise study results. Generally speaking, patients are less adherent to treatment when they feel well, such as when taking prophylactic treatment, than they are for a symptomatic condition [14], [15]. We report on non-adherence and loss to follow-up in 1,995 HIV-infected adults enrolled in a 3-year prophylaxis trial in two Botswana cities. Based upon the pharmacy refill criterion for adherence (i.e.,≥80% attendance to study medication refill visits), despite a prolonged 36-month period of prophylaxis, participants in our trial had a 78% adherence rate which is similar to the 69–86% rates reported by other clinical trials that provided 6 months of IPT [16]–[18]. Using the urine isoniazid metabolite criterion as an assessment of adherence, participants in the 36-month Botswana study had a 74–80% adherence rate which is comparable to the 45–80% rates reported in two other IPT clinical trials of shorter duration [16], [18]. We conducted a case-control sub-study in which we interviewed 83 participants who had been lost to follow-up and 127 participants who were non-adherent to study medication but still attending visits and compared their demographics and responses to specific questions with those of 252 adherent participants. Factors associated with non-adherence or loss to follow-up included age, male sex and difficulties with the regimen. Additional information from participants identified side effects and competing commitments such as work, relocation, and the belief that they had completed the study as contributions to non-adherence or loss to follow-up.

In the case-control sub-study, men had twice and thrice the odds of women to be non-adherent or lost to follow-up, respectively. While we were unable to identify published studies examining adherence to prophylactic therapy in large numbers of HIV-infected adults in sub-Saharan Africa, adherence to ART has been assessed in such populations. Among enrollees of ART programs in Côte d'Ivoire and Kenya, men were more likely than women to be lost to follow-up [19], [20]. In South Africa, based upon pharmacy records, women were more likely than men (54% vs 49%, P <0.001) to take ≥80% of their ART [21], although this phenomenon is not limited to South Africa [22]. While reasons for the increased risk of non-adherence in African men attending public health clinics are poorly understood, employment circumstances and frequency and amount of alcohol use are two possible explanations. The employment rate in men in the Botswana trial was higher than women (82% vs. 63%), but the type of employment and the work hours of men may have made adherence more difficult. Indeed interview and focus group participants noted that taking time off from work for study visits was often difficult. Widespread heavy alcohol use has been reported in Botswana communities, particularly in men (31% vs. 17% in women) [23] and was paralleled by observations from the IPT trial (data not shown). Reasons why HIV-infected men are more likely to be non-adherent need to be identified in future studies and should guide approaches taken to improve their adherence.

Non-adherent participants had more than three times the odds of reporting difficulties with the regimen or not knowing if they had difficulties than controls (rates reported were 25% in non-adherent cases vs. 7% in controls) and although not significant in the multivariable model, participants lost to follow-up also had higher rates (16%) of reporting these difficulties or not knowing. Some discussion group and interview participants reported that the study medication made them sick, lose weight, feel dizzy or tired or made their body ache. It is well established that side-effects can greatly influence an individual's willingness to adhere to therapy [24], [25]. Based upon trial records, none of these individuals suffered adverse effects of the study medication. However it is possible that they may have had mild but irritating symptoms which they did not report to the study staff. This may represent reporting bias in that patients are more likely to self-report adverse effects of a medication than clinicians are to record them [26] or it may represent an excuse for stopping the medication. Younger age was also associated with non-adherent participants as has been commonly reported in the medical literature and also for HIV-infected persons [27]–[29].

Some associations that were significant in bivariate analysis were not significant in the adjusted analyses but bear mentioning:distance from clinic, alcohol use, level of education, and understanding the requirements of the study. All four factors have been reported as barriers to adherence [30], [22], [31], [32]. It may be that these factors were not independently associated with non-adherent or lost to follow-up participants in our study because of small numbers, differences in question comprehension, or confounding with other variables. For instance, alcohol use may not have been significant in the adjusted analysis because the question did not ask about the frequency and quantity of alcohol consumed. Notably, a relatively large proportion of both groups (17% non-adherents and 13% lost to follow-up) reported stopping the medication because they believed they had completed the study. This may indicate that either nursing staff did not adequately explain study requirements to the participants or that the participant did not fully comprehend what was being explained. Alternatively, as posited by nursing staff, some cases may have used this option as an excuse because all participants were required to pass a comprehension test before enrollment.

Our study had several potential limitations. First, data were cross-sectional, so we cannot state that the significant factors led to a participant becoming non-adherent or lost to follow-up. Second, responses to survey questions were self-reported which may have resulted in some bias in answering questions about non-adherence, although we believe this was minimized since the study was conducted by an independent contractor and not members of the clinic staff. Third, the clinical trial staff did not have control over the independent administration of ART to study participants during their changing disease course over the 3-year study. As HIV-infected persons widely regard ART as life-sustaining and as ART consists of many additional pills and also have significant adverse effects, they may have complicated adherence to study medication. Fourth, we did not assess the personalities of the participants which could have potentially been a factor in their adherence to study medication. Fifth, because we were unable to contact 197 participants who had voluntarily withdrawn at the time the sub-study was begun, the lack of information about them may have biased our sub-study. Finally, all of the relevant and important variables may not have been captured in our survey (e.g., peer or family social support). The strengths of this study are the inclusion of group discussions and interviews to supplement the survey data, clear definitions of non-adherence and loss to follow-up, and the very high survey response rate.

Studies have shown that participant satisfaction with the clinic staff can contribute to adherence [32]. In our study, discussion group and interview participants overwhelmingly praised their treatment from the clinic staff in a number of areas. Moreover, the majority of cases responding to the survey agreed that they were treated well by the clinic staff to the degree that the variables could not be included in the multiple regression model because the number of negative responses was too small. Suggestions from the discussion group and interview participants as to what would improve their experience ranged from adding weekend clinic hours to adding more clinics in different locations, but none related to the way they were treated by clinic staff. For future clinical trials, potential interventions that might improve adherence in trial participants, even for non-modifiable factors like sex, include targeting health education for men, reducing barriers (e.g. offering more clinics in different locations if possible and weekend hours, making visits brief and convenient), clarifying study expectations, educating employers about HIV/AIDS to help reduce stigma in the workplace, and encouraging employers to support employee health.